xref: /linux/kernel/cgroup/rstat.c (revision a1ff5a7d78a036d6c2178ee5acd6ba4946243800)
1 // SPDX-License-Identifier: GPL-2.0-only
2 #include "cgroup-internal.h"
3 
4 #include <linux/sched/cputime.h>
5 
6 #include <linux/bpf.h>
7 #include <linux/btf.h>
8 #include <linux/btf_ids.h>
9 
10 #include <trace/events/cgroup.h>
11 
12 static DEFINE_SPINLOCK(cgroup_rstat_lock);
13 static DEFINE_PER_CPU(raw_spinlock_t, cgroup_rstat_cpu_lock);
14 
15 static void cgroup_base_stat_flush(struct cgroup *cgrp, int cpu);
16 
cgroup_rstat_cpu(struct cgroup * cgrp,int cpu)17 static struct cgroup_rstat_cpu *cgroup_rstat_cpu(struct cgroup *cgrp, int cpu)
18 {
19 	return per_cpu_ptr(cgrp->rstat_cpu, cpu);
20 }
21 
22 /*
23  * Helper functions for rstat per CPU lock (cgroup_rstat_cpu_lock).
24  *
25  * This makes it easier to diagnose locking issues and contention in
26  * production environments. The parameter @fast_path determine the
27  * tracepoints being added, allowing us to diagnose "flush" related
28  * operations without handling high-frequency fast-path "update" events.
29  */
30 static __always_inline
_cgroup_rstat_cpu_lock(raw_spinlock_t * cpu_lock,int cpu,struct cgroup * cgrp,const bool fast_path)31 unsigned long _cgroup_rstat_cpu_lock(raw_spinlock_t *cpu_lock, int cpu,
32 				     struct cgroup *cgrp, const bool fast_path)
33 {
34 	unsigned long flags;
35 	bool contended;
36 
37 	/*
38 	 * The _irqsave() is needed because cgroup_rstat_lock is
39 	 * spinlock_t which is a sleeping lock on PREEMPT_RT. Acquiring
40 	 * this lock with the _irq() suffix only disables interrupts on
41 	 * a non-PREEMPT_RT kernel. The raw_spinlock_t below disables
42 	 * interrupts on both configurations. The _irqsave() ensures
43 	 * that interrupts are always disabled and later restored.
44 	 */
45 	contended = !raw_spin_trylock_irqsave(cpu_lock, flags);
46 	if (contended) {
47 		if (fast_path)
48 			trace_cgroup_rstat_cpu_lock_contended_fastpath(cgrp, cpu, contended);
49 		else
50 			trace_cgroup_rstat_cpu_lock_contended(cgrp, cpu, contended);
51 
52 		raw_spin_lock_irqsave(cpu_lock, flags);
53 	}
54 
55 	if (fast_path)
56 		trace_cgroup_rstat_cpu_locked_fastpath(cgrp, cpu, contended);
57 	else
58 		trace_cgroup_rstat_cpu_locked(cgrp, cpu, contended);
59 
60 	return flags;
61 }
62 
63 static __always_inline
_cgroup_rstat_cpu_unlock(raw_spinlock_t * cpu_lock,int cpu,struct cgroup * cgrp,unsigned long flags,const bool fast_path)64 void _cgroup_rstat_cpu_unlock(raw_spinlock_t *cpu_lock, int cpu,
65 			      struct cgroup *cgrp, unsigned long flags,
66 			      const bool fast_path)
67 {
68 	if (fast_path)
69 		trace_cgroup_rstat_cpu_unlock_fastpath(cgrp, cpu, false);
70 	else
71 		trace_cgroup_rstat_cpu_unlock(cgrp, cpu, false);
72 
73 	raw_spin_unlock_irqrestore(cpu_lock, flags);
74 }
75 
76 /**
77  * cgroup_rstat_updated - keep track of updated rstat_cpu
78  * @cgrp: target cgroup
79  * @cpu: cpu on which rstat_cpu was updated
80  *
81  * @cgrp's rstat_cpu on @cpu was updated.  Put it on the parent's matching
82  * rstat_cpu->updated_children list.  See the comment on top of
83  * cgroup_rstat_cpu definition for details.
84  */
cgroup_rstat_updated(struct cgroup * cgrp,int cpu)85 __bpf_kfunc void cgroup_rstat_updated(struct cgroup *cgrp, int cpu)
86 {
87 	raw_spinlock_t *cpu_lock = per_cpu_ptr(&cgroup_rstat_cpu_lock, cpu);
88 	unsigned long flags;
89 
90 	/*
91 	 * Speculative already-on-list test. This may race leading to
92 	 * temporary inaccuracies, which is fine.
93 	 *
94 	 * Because @parent's updated_children is terminated with @parent
95 	 * instead of NULL, we can tell whether @cgrp is on the list by
96 	 * testing the next pointer for NULL.
97 	 */
98 	if (data_race(cgroup_rstat_cpu(cgrp, cpu)->updated_next))
99 		return;
100 
101 	flags = _cgroup_rstat_cpu_lock(cpu_lock, cpu, cgrp, true);
102 
103 	/* put @cgrp and all ancestors on the corresponding updated lists */
104 	while (true) {
105 		struct cgroup_rstat_cpu *rstatc = cgroup_rstat_cpu(cgrp, cpu);
106 		struct cgroup *parent = cgroup_parent(cgrp);
107 		struct cgroup_rstat_cpu *prstatc;
108 
109 		/*
110 		 * Both additions and removals are bottom-up.  If a cgroup
111 		 * is already in the tree, all ancestors are.
112 		 */
113 		if (rstatc->updated_next)
114 			break;
115 
116 		/* Root has no parent to link it to, but mark it busy */
117 		if (!parent) {
118 			rstatc->updated_next = cgrp;
119 			break;
120 		}
121 
122 		prstatc = cgroup_rstat_cpu(parent, cpu);
123 		rstatc->updated_next = prstatc->updated_children;
124 		prstatc->updated_children = cgrp;
125 
126 		cgrp = parent;
127 	}
128 
129 	_cgroup_rstat_cpu_unlock(cpu_lock, cpu, cgrp, flags, true);
130 }
131 
132 /**
133  * cgroup_rstat_push_children - push children cgroups into the given list
134  * @head: current head of the list (= subtree root)
135  * @child: first child of the root
136  * @cpu: target cpu
137  * Return: A new singly linked list of cgroups to be flush
138  *
139  * Iteratively traverse down the cgroup_rstat_cpu updated tree level by
140  * level and push all the parents first before their next level children
141  * into a singly linked list built from the tail backward like "pushing"
142  * cgroups into a stack. The root is pushed by the caller.
143  */
cgroup_rstat_push_children(struct cgroup * head,struct cgroup * child,int cpu)144 static struct cgroup *cgroup_rstat_push_children(struct cgroup *head,
145 						 struct cgroup *child, int cpu)
146 {
147 	struct cgroup *chead = child;	/* Head of child cgroup level */
148 	struct cgroup *ghead = NULL;	/* Head of grandchild cgroup level */
149 	struct cgroup *parent, *grandchild;
150 	struct cgroup_rstat_cpu *crstatc;
151 
152 	child->rstat_flush_next = NULL;
153 
154 next_level:
155 	while (chead) {
156 		child = chead;
157 		chead = child->rstat_flush_next;
158 		parent = cgroup_parent(child);
159 
160 		/* updated_next is parent cgroup terminated */
161 		while (child != parent) {
162 			child->rstat_flush_next = head;
163 			head = child;
164 			crstatc = cgroup_rstat_cpu(child, cpu);
165 			grandchild = crstatc->updated_children;
166 			if (grandchild != child) {
167 				/* Push the grand child to the next level */
168 				crstatc->updated_children = child;
169 				grandchild->rstat_flush_next = ghead;
170 				ghead = grandchild;
171 			}
172 			child = crstatc->updated_next;
173 			crstatc->updated_next = NULL;
174 		}
175 	}
176 
177 	if (ghead) {
178 		chead = ghead;
179 		ghead = NULL;
180 		goto next_level;
181 	}
182 	return head;
183 }
184 
185 /**
186  * cgroup_rstat_updated_list - return a list of updated cgroups to be flushed
187  * @root: root of the cgroup subtree to traverse
188  * @cpu: target cpu
189  * Return: A singly linked list of cgroups to be flushed
190  *
191  * Walks the updated rstat_cpu tree on @cpu from @root.  During traversal,
192  * each returned cgroup is unlinked from the updated tree.
193  *
194  * The only ordering guarantee is that, for a parent and a child pair
195  * covered by a given traversal, the child is before its parent in
196  * the list.
197  *
198  * Note that updated_children is self terminated and points to a list of
199  * child cgroups if not empty. Whereas updated_next is like a sibling link
200  * within the children list and terminated by the parent cgroup. An exception
201  * here is the cgroup root whose updated_next can be self terminated.
202  */
cgroup_rstat_updated_list(struct cgroup * root,int cpu)203 static struct cgroup *cgroup_rstat_updated_list(struct cgroup *root, int cpu)
204 {
205 	raw_spinlock_t *cpu_lock = per_cpu_ptr(&cgroup_rstat_cpu_lock, cpu);
206 	struct cgroup_rstat_cpu *rstatc = cgroup_rstat_cpu(root, cpu);
207 	struct cgroup *head = NULL, *parent, *child;
208 	unsigned long flags;
209 
210 	flags = _cgroup_rstat_cpu_lock(cpu_lock, cpu, root, false);
211 
212 	/* Return NULL if this subtree is not on-list */
213 	if (!rstatc->updated_next)
214 		goto unlock_ret;
215 
216 	/*
217 	 * Unlink @root from its parent. As the updated_children list is
218 	 * singly linked, we have to walk it to find the removal point.
219 	 */
220 	parent = cgroup_parent(root);
221 	if (parent) {
222 		struct cgroup_rstat_cpu *prstatc;
223 		struct cgroup **nextp;
224 
225 		prstatc = cgroup_rstat_cpu(parent, cpu);
226 		nextp = &prstatc->updated_children;
227 		while (*nextp != root) {
228 			struct cgroup_rstat_cpu *nrstatc;
229 
230 			nrstatc = cgroup_rstat_cpu(*nextp, cpu);
231 			WARN_ON_ONCE(*nextp == parent);
232 			nextp = &nrstatc->updated_next;
233 		}
234 		*nextp = rstatc->updated_next;
235 	}
236 
237 	rstatc->updated_next = NULL;
238 
239 	/* Push @root to the list first before pushing the children */
240 	head = root;
241 	root->rstat_flush_next = NULL;
242 	child = rstatc->updated_children;
243 	rstatc->updated_children = root;
244 	if (child != root)
245 		head = cgroup_rstat_push_children(head, child, cpu);
246 unlock_ret:
247 	_cgroup_rstat_cpu_unlock(cpu_lock, cpu, root, flags, false);
248 	return head;
249 }
250 
251 /*
252  * A hook for bpf stat collectors to attach to and flush their stats.
253  * Together with providing bpf kfuncs for cgroup_rstat_updated() and
254  * cgroup_rstat_flush(), this enables a complete workflow where bpf progs that
255  * collect cgroup stats can integrate with rstat for efficient flushing.
256  *
257  * A static noinline declaration here could cause the compiler to optimize away
258  * the function. A global noinline declaration will keep the definition, but may
259  * optimize away the callsite. Therefore, __weak is needed to ensure that the
260  * call is still emitted, by telling the compiler that we don't know what the
261  * function might eventually be.
262  */
263 
264 __bpf_hook_start();
265 
bpf_rstat_flush(struct cgroup * cgrp,struct cgroup * parent,int cpu)266 __weak noinline void bpf_rstat_flush(struct cgroup *cgrp,
267 				     struct cgroup *parent, int cpu)
268 {
269 }
270 
271 __bpf_hook_end();
272 
273 /*
274  * Helper functions for locking cgroup_rstat_lock.
275  *
276  * This makes it easier to diagnose locking issues and contention in
277  * production environments.  The parameter @cpu_in_loop indicate lock
278  * was released and re-taken when collection data from the CPUs. The
279  * value -1 is used when obtaining the main lock else this is the CPU
280  * number processed last.
281  */
__cgroup_rstat_lock(struct cgroup * cgrp,int cpu_in_loop)282 static inline void __cgroup_rstat_lock(struct cgroup *cgrp, int cpu_in_loop)
283 	__acquires(&cgroup_rstat_lock)
284 {
285 	bool contended;
286 
287 	contended = !spin_trylock_irq(&cgroup_rstat_lock);
288 	if (contended) {
289 		trace_cgroup_rstat_lock_contended(cgrp, cpu_in_loop, contended);
290 		spin_lock_irq(&cgroup_rstat_lock);
291 	}
292 	trace_cgroup_rstat_locked(cgrp, cpu_in_loop, contended);
293 }
294 
__cgroup_rstat_unlock(struct cgroup * cgrp,int cpu_in_loop)295 static inline void __cgroup_rstat_unlock(struct cgroup *cgrp, int cpu_in_loop)
296 	__releases(&cgroup_rstat_lock)
297 {
298 	trace_cgroup_rstat_unlock(cgrp, cpu_in_loop, false);
299 	spin_unlock_irq(&cgroup_rstat_lock);
300 }
301 
302 /* see cgroup_rstat_flush() */
cgroup_rstat_flush_locked(struct cgroup * cgrp)303 static void cgroup_rstat_flush_locked(struct cgroup *cgrp)
304 	__releases(&cgroup_rstat_lock) __acquires(&cgroup_rstat_lock)
305 {
306 	int cpu;
307 
308 	lockdep_assert_held(&cgroup_rstat_lock);
309 
310 	for_each_possible_cpu(cpu) {
311 		struct cgroup *pos = cgroup_rstat_updated_list(cgrp, cpu);
312 
313 		for (; pos; pos = pos->rstat_flush_next) {
314 			struct cgroup_subsys_state *css;
315 
316 			cgroup_base_stat_flush(pos, cpu);
317 			bpf_rstat_flush(pos, cgroup_parent(pos), cpu);
318 
319 			rcu_read_lock();
320 			list_for_each_entry_rcu(css, &pos->rstat_css_list,
321 						rstat_css_node)
322 				css->ss->css_rstat_flush(css, cpu);
323 			rcu_read_unlock();
324 		}
325 
326 		/* play nice and yield if necessary */
327 		if (need_resched() || spin_needbreak(&cgroup_rstat_lock)) {
328 			__cgroup_rstat_unlock(cgrp, cpu);
329 			if (!cond_resched())
330 				cpu_relax();
331 			__cgroup_rstat_lock(cgrp, cpu);
332 		}
333 	}
334 }
335 
336 /**
337  * cgroup_rstat_flush - flush stats in @cgrp's subtree
338  * @cgrp: target cgroup
339  *
340  * Collect all per-cpu stats in @cgrp's subtree into the global counters
341  * and propagate them upwards.  After this function returns, all cgroups in
342  * the subtree have up-to-date ->stat.
343  *
344  * This also gets all cgroups in the subtree including @cgrp off the
345  * ->updated_children lists.
346  *
347  * This function may block.
348  */
cgroup_rstat_flush(struct cgroup * cgrp)349 __bpf_kfunc void cgroup_rstat_flush(struct cgroup *cgrp)
350 {
351 	might_sleep();
352 
353 	__cgroup_rstat_lock(cgrp, -1);
354 	cgroup_rstat_flush_locked(cgrp);
355 	__cgroup_rstat_unlock(cgrp, -1);
356 }
357 
358 /**
359  * cgroup_rstat_flush_hold - flush stats in @cgrp's subtree and hold
360  * @cgrp: target cgroup
361  *
362  * Flush stats in @cgrp's subtree and prevent further flushes.  Must be
363  * paired with cgroup_rstat_flush_release().
364  *
365  * This function may block.
366  */
cgroup_rstat_flush_hold(struct cgroup * cgrp)367 void cgroup_rstat_flush_hold(struct cgroup *cgrp)
368 	__acquires(&cgroup_rstat_lock)
369 {
370 	might_sleep();
371 	__cgroup_rstat_lock(cgrp, -1);
372 	cgroup_rstat_flush_locked(cgrp);
373 }
374 
375 /**
376  * cgroup_rstat_flush_release - release cgroup_rstat_flush_hold()
377  * @cgrp: cgroup used by tracepoint
378  */
cgroup_rstat_flush_release(struct cgroup * cgrp)379 void cgroup_rstat_flush_release(struct cgroup *cgrp)
380 	__releases(&cgroup_rstat_lock)
381 {
382 	__cgroup_rstat_unlock(cgrp, -1);
383 }
384 
cgroup_rstat_init(struct cgroup * cgrp)385 int cgroup_rstat_init(struct cgroup *cgrp)
386 {
387 	int cpu;
388 
389 	/* the root cgrp has rstat_cpu preallocated */
390 	if (!cgrp->rstat_cpu) {
391 		cgrp->rstat_cpu = alloc_percpu(struct cgroup_rstat_cpu);
392 		if (!cgrp->rstat_cpu)
393 			return -ENOMEM;
394 	}
395 
396 	/* ->updated_children list is self terminated */
397 	for_each_possible_cpu(cpu) {
398 		struct cgroup_rstat_cpu *rstatc = cgroup_rstat_cpu(cgrp, cpu);
399 
400 		rstatc->updated_children = cgrp;
401 		u64_stats_init(&rstatc->bsync);
402 	}
403 
404 	return 0;
405 }
406 
cgroup_rstat_exit(struct cgroup * cgrp)407 void cgroup_rstat_exit(struct cgroup *cgrp)
408 {
409 	int cpu;
410 
411 	cgroup_rstat_flush(cgrp);
412 
413 	/* sanity check */
414 	for_each_possible_cpu(cpu) {
415 		struct cgroup_rstat_cpu *rstatc = cgroup_rstat_cpu(cgrp, cpu);
416 
417 		if (WARN_ON_ONCE(rstatc->updated_children != cgrp) ||
418 		    WARN_ON_ONCE(rstatc->updated_next))
419 			return;
420 	}
421 
422 	free_percpu(cgrp->rstat_cpu);
423 	cgrp->rstat_cpu = NULL;
424 }
425 
cgroup_rstat_boot(void)426 void __init cgroup_rstat_boot(void)
427 {
428 	int cpu;
429 
430 	for_each_possible_cpu(cpu)
431 		raw_spin_lock_init(per_cpu_ptr(&cgroup_rstat_cpu_lock, cpu));
432 }
433 
434 /*
435  * Functions for cgroup basic resource statistics implemented on top of
436  * rstat.
437  */
cgroup_base_stat_add(struct cgroup_base_stat * dst_bstat,struct cgroup_base_stat * src_bstat)438 static void cgroup_base_stat_add(struct cgroup_base_stat *dst_bstat,
439 				 struct cgroup_base_stat *src_bstat)
440 {
441 	dst_bstat->cputime.utime += src_bstat->cputime.utime;
442 	dst_bstat->cputime.stime += src_bstat->cputime.stime;
443 	dst_bstat->cputime.sum_exec_runtime += src_bstat->cputime.sum_exec_runtime;
444 #ifdef CONFIG_SCHED_CORE
445 	dst_bstat->forceidle_sum += src_bstat->forceidle_sum;
446 #endif
447 }
448 
cgroup_base_stat_sub(struct cgroup_base_stat * dst_bstat,struct cgroup_base_stat * src_bstat)449 static void cgroup_base_stat_sub(struct cgroup_base_stat *dst_bstat,
450 				 struct cgroup_base_stat *src_bstat)
451 {
452 	dst_bstat->cputime.utime -= src_bstat->cputime.utime;
453 	dst_bstat->cputime.stime -= src_bstat->cputime.stime;
454 	dst_bstat->cputime.sum_exec_runtime -= src_bstat->cputime.sum_exec_runtime;
455 #ifdef CONFIG_SCHED_CORE
456 	dst_bstat->forceidle_sum -= src_bstat->forceidle_sum;
457 #endif
458 }
459 
cgroup_base_stat_flush(struct cgroup * cgrp,int cpu)460 static void cgroup_base_stat_flush(struct cgroup *cgrp, int cpu)
461 {
462 	struct cgroup_rstat_cpu *rstatc = cgroup_rstat_cpu(cgrp, cpu);
463 	struct cgroup *parent = cgroup_parent(cgrp);
464 	struct cgroup_rstat_cpu *prstatc;
465 	struct cgroup_base_stat delta;
466 	unsigned seq;
467 
468 	/* Root-level stats are sourced from system-wide CPU stats */
469 	if (!parent)
470 		return;
471 
472 	/* fetch the current per-cpu values */
473 	do {
474 		seq = __u64_stats_fetch_begin(&rstatc->bsync);
475 		delta = rstatc->bstat;
476 	} while (__u64_stats_fetch_retry(&rstatc->bsync, seq));
477 
478 	/* propagate per-cpu delta to cgroup and per-cpu global statistics */
479 	cgroup_base_stat_sub(&delta, &rstatc->last_bstat);
480 	cgroup_base_stat_add(&cgrp->bstat, &delta);
481 	cgroup_base_stat_add(&rstatc->last_bstat, &delta);
482 	cgroup_base_stat_add(&rstatc->subtree_bstat, &delta);
483 
484 	/* propagate cgroup and per-cpu global delta to parent (unless that's root) */
485 	if (cgroup_parent(parent)) {
486 		delta = cgrp->bstat;
487 		cgroup_base_stat_sub(&delta, &cgrp->last_bstat);
488 		cgroup_base_stat_add(&parent->bstat, &delta);
489 		cgroup_base_stat_add(&cgrp->last_bstat, &delta);
490 
491 		delta = rstatc->subtree_bstat;
492 		prstatc = cgroup_rstat_cpu(parent, cpu);
493 		cgroup_base_stat_sub(&delta, &rstatc->last_subtree_bstat);
494 		cgroup_base_stat_add(&prstatc->subtree_bstat, &delta);
495 		cgroup_base_stat_add(&rstatc->last_subtree_bstat, &delta);
496 	}
497 }
498 
499 static struct cgroup_rstat_cpu *
cgroup_base_stat_cputime_account_begin(struct cgroup * cgrp,unsigned long * flags)500 cgroup_base_stat_cputime_account_begin(struct cgroup *cgrp, unsigned long *flags)
501 {
502 	struct cgroup_rstat_cpu *rstatc;
503 
504 	rstatc = get_cpu_ptr(cgrp->rstat_cpu);
505 	*flags = u64_stats_update_begin_irqsave(&rstatc->bsync);
506 	return rstatc;
507 }
508 
cgroup_base_stat_cputime_account_end(struct cgroup * cgrp,struct cgroup_rstat_cpu * rstatc,unsigned long flags)509 static void cgroup_base_stat_cputime_account_end(struct cgroup *cgrp,
510 						 struct cgroup_rstat_cpu *rstatc,
511 						 unsigned long flags)
512 {
513 	u64_stats_update_end_irqrestore(&rstatc->bsync, flags);
514 	cgroup_rstat_updated(cgrp, smp_processor_id());
515 	put_cpu_ptr(rstatc);
516 }
517 
__cgroup_account_cputime(struct cgroup * cgrp,u64 delta_exec)518 void __cgroup_account_cputime(struct cgroup *cgrp, u64 delta_exec)
519 {
520 	struct cgroup_rstat_cpu *rstatc;
521 	unsigned long flags;
522 
523 	rstatc = cgroup_base_stat_cputime_account_begin(cgrp, &flags);
524 	rstatc->bstat.cputime.sum_exec_runtime += delta_exec;
525 	cgroup_base_stat_cputime_account_end(cgrp, rstatc, flags);
526 }
527 
__cgroup_account_cputime_field(struct cgroup * cgrp,enum cpu_usage_stat index,u64 delta_exec)528 void __cgroup_account_cputime_field(struct cgroup *cgrp,
529 				    enum cpu_usage_stat index, u64 delta_exec)
530 {
531 	struct cgroup_rstat_cpu *rstatc;
532 	unsigned long flags;
533 
534 	rstatc = cgroup_base_stat_cputime_account_begin(cgrp, &flags);
535 
536 	switch (index) {
537 	case CPUTIME_USER:
538 	case CPUTIME_NICE:
539 		rstatc->bstat.cputime.utime += delta_exec;
540 		break;
541 	case CPUTIME_SYSTEM:
542 	case CPUTIME_IRQ:
543 	case CPUTIME_SOFTIRQ:
544 		rstatc->bstat.cputime.stime += delta_exec;
545 		break;
546 #ifdef CONFIG_SCHED_CORE
547 	case CPUTIME_FORCEIDLE:
548 		rstatc->bstat.forceidle_sum += delta_exec;
549 		break;
550 #endif
551 	default:
552 		break;
553 	}
554 
555 	cgroup_base_stat_cputime_account_end(cgrp, rstatc, flags);
556 }
557 
558 /*
559  * compute the cputime for the root cgroup by getting the per cpu data
560  * at a global level, then categorizing the fields in a manner consistent
561  * with how it is done by __cgroup_account_cputime_field for each bit of
562  * cpu time attributed to a cgroup.
563  */
root_cgroup_cputime(struct cgroup_base_stat * bstat)564 static void root_cgroup_cputime(struct cgroup_base_stat *bstat)
565 {
566 	struct task_cputime *cputime = &bstat->cputime;
567 	int i;
568 
569 	memset(bstat, 0, sizeof(*bstat));
570 	for_each_possible_cpu(i) {
571 		struct kernel_cpustat kcpustat;
572 		u64 *cpustat = kcpustat.cpustat;
573 		u64 user = 0;
574 		u64 sys = 0;
575 
576 		kcpustat_cpu_fetch(&kcpustat, i);
577 
578 		user += cpustat[CPUTIME_USER];
579 		user += cpustat[CPUTIME_NICE];
580 		cputime->utime += user;
581 
582 		sys += cpustat[CPUTIME_SYSTEM];
583 		sys += cpustat[CPUTIME_IRQ];
584 		sys += cpustat[CPUTIME_SOFTIRQ];
585 		cputime->stime += sys;
586 
587 		cputime->sum_exec_runtime += user;
588 		cputime->sum_exec_runtime += sys;
589 		cputime->sum_exec_runtime += cpustat[CPUTIME_STEAL];
590 
591 #ifdef CONFIG_SCHED_CORE
592 		bstat->forceidle_sum += cpustat[CPUTIME_FORCEIDLE];
593 #endif
594 	}
595 }
596 
597 
cgroup_force_idle_show(struct seq_file * seq,struct cgroup_base_stat * bstat)598 static void cgroup_force_idle_show(struct seq_file *seq, struct cgroup_base_stat *bstat)
599 {
600 #ifdef CONFIG_SCHED_CORE
601 	u64 forceidle_time = bstat->forceidle_sum;
602 
603 	do_div(forceidle_time, NSEC_PER_USEC);
604 	seq_printf(seq, "core_sched.force_idle_usec %llu\n", forceidle_time);
605 #endif
606 }
607 
cgroup_base_stat_cputime_show(struct seq_file * seq)608 void cgroup_base_stat_cputime_show(struct seq_file *seq)
609 {
610 	struct cgroup *cgrp = seq_css(seq)->cgroup;
611 	u64 usage, utime, stime;
612 
613 	if (cgroup_parent(cgrp)) {
614 		cgroup_rstat_flush_hold(cgrp);
615 		usage = cgrp->bstat.cputime.sum_exec_runtime;
616 		cputime_adjust(&cgrp->bstat.cputime, &cgrp->prev_cputime,
617 			       &utime, &stime);
618 		cgroup_rstat_flush_release(cgrp);
619 	} else {
620 		/* cgrp->bstat of root is not actually used, reuse it */
621 		root_cgroup_cputime(&cgrp->bstat);
622 		usage = cgrp->bstat.cputime.sum_exec_runtime;
623 		utime = cgrp->bstat.cputime.utime;
624 		stime = cgrp->bstat.cputime.stime;
625 	}
626 
627 	do_div(usage, NSEC_PER_USEC);
628 	do_div(utime, NSEC_PER_USEC);
629 	do_div(stime, NSEC_PER_USEC);
630 
631 	seq_printf(seq, "usage_usec %llu\n"
632 		   "user_usec %llu\n"
633 		   "system_usec %llu\n",
634 		   usage, utime, stime);
635 
636 	cgroup_force_idle_show(seq, &cgrp->bstat);
637 }
638 
639 /* Add bpf kfuncs for cgroup_rstat_updated() and cgroup_rstat_flush() */
640 BTF_KFUNCS_START(bpf_rstat_kfunc_ids)
641 BTF_ID_FLAGS(func, cgroup_rstat_updated)
642 BTF_ID_FLAGS(func, cgroup_rstat_flush, KF_SLEEPABLE)
643 BTF_KFUNCS_END(bpf_rstat_kfunc_ids)
644 
645 static const struct btf_kfunc_id_set bpf_rstat_kfunc_set = {
646 	.owner          = THIS_MODULE,
647 	.set            = &bpf_rstat_kfunc_ids,
648 };
649 
bpf_rstat_kfunc_init(void)650 static int __init bpf_rstat_kfunc_init(void)
651 {
652 	return register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING,
653 					 &bpf_rstat_kfunc_set);
654 }
655 late_initcall(bpf_rstat_kfunc_init);
656